Hand action recognition method based on deep reinforcement learning and double-feature double-motor neural network

The invention relates to the technical field of intelligent production line man-machine cooperation, in particular to a hand action recognition method based on deep reinforcement learning and a double-feature double-motor neural network. The method comprises the following steps: firstly, training a double-feature double-motor neural network, then constructing a deep reinforcement learning selection frame network, selecting key frames according to the constructed selection frame network, inputting the key frames into the double-feature double-motor neural network, and carrying out retraining to obtain a new network model; and finally, inputting the hand skeleton sequence of the test set into the deep reinforcement learning selection frame network, selecting a key frame, inputting the selected key frame into the new network model, and finally obtaining a classification result of the hand skeleton action sequence in the test set. According to the method, key frames are selected through the Markov decision proces